Science Question Answering on ScienceQA Image (test)
80.4AccuracyQwen
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QwenBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=100%2025.08 | 80.4 | 100 | |
| HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.08 | 80.2 | 97.1 | |
| HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.08 | 80.2 | 97.1 | |
| VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.08 | 80.1 | 97.7 | |
| VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.08 | 80 | 96.2 | |
| HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.08 | 79.9 | 98.7 | |
| FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.08 | 79.6 | 95.9 | |
| VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.08 | 79.5 | 91.5 | |
| HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.08 | 79.4 | 98.7 | |
| HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.08 | 79.4 | 93 | |
| FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.08 | 79.3 | 97.4 | |
| FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.08 | 79.3 | 86.4 | |
| HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.08 | 79.1 | 93 |